dreaminglisted
Install: claude install-skill ymurat/academic-agent-skills
# Dreaming
Propose-then-apply memory consolidation for agents whose durable store is typed memory (for example `profile`, `journal`, `shared_profile`, `shared_journal`), not a `MEMORY.md` file.
The pattern follows OpenClaw-style sleep-time consolidation (harvest, score, reflect) and Hermes-style staging: live memory is never mutated during harvest or review. A later apply happens only after the user explicitly approves named items.
## The rule
Write live memory only after the user explicitly approves specific proposed facts. A nightly run, a request to "dream" or "scan memory", or a review of candidates is not approval. A bare "write memory" or "apply" that does not name the items is not approval either: apply only the items they accepted.
If the user said "remember that X" in the current turn, write that fact with the normal memory tool and stop. This skill is for consolidation, not for a single explicit remember request.
## Memory types
Some agents use a typed memory store with kinds such as:
- `profile`: lasting facts about the user and how this assistant should work with them. Keep it small.
- `journal`: dated history. Only recent entries stay in view.
- `shared_profile`: lasting facts every assistant this user runs should know (name, timezone, standing preferences that apply across bots).
- `shared_journal`: dated history every assistant should know.
Other agents maintain all their memory in a `MEMORY.md` file with sections that map to the same concepts (for exa